]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/dir.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * nfs directory handling functions | |
7 | * | |
8 | * 10 Apr 1996 Added silly rename for unlink --okir | |
9 | * 28 Sep 1996 Improved directory cache --okir | |
10 | * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de | |
11 | * Re-implemented silly rename for unlink, newly implemented | |
12 | * silly rename for nfs_rename() following the suggestions | |
13 | * of Olaf Kirch (okir) found in this file. | |
14 | * Following Linus comments on my original hack, this version | |
15 | * depends only on the dcache stuff and doesn't touch the inode | |
16 | * layer (iput() and friends). | |
17 | * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/stat.h> | |
23 | #include <linux/fcntl.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
29 | #include <linux/nfs_fs.h> | |
30 | #include <linux/nfs_mount.h> | |
31 | #include <linux/pagemap.h> | |
873101b3 | 32 | #include <linux/pagevec.h> |
1da177e4 | 33 | #include <linux/namei.h> |
54ceac45 | 34 | #include <linux/mount.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
56e4ebf8 | 36 | #include <linux/vmalloc.h> |
04e4bd1c | 37 | #include <linux/kmemleak.h> |
1da177e4 LT |
38 | |
39 | #include "delegation.h" | |
91d5b470 | 40 | #include "iostat.h" |
4c30d56e | 41 | #include "internal.h" |
cd9a1c0e | 42 | #include "fscache.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | /* #define NFS_DEBUG_VERBOSE 1 */ |
45 | ||
46 | static int nfs_opendir(struct inode *, struct file *); | |
47 | static int nfs_readdir(struct file *, void *, filldir_t); | |
48 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
49 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
50 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
51 | static int nfs_rmdir(struct inode *, struct dentry *); | |
52 | static int nfs_unlink(struct inode *, struct dentry *); | |
53 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
54 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
55 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
56 | static int nfs_rename(struct inode *, struct dentry *, | |
57 | struct inode *, struct dentry *); | |
7ea80859 | 58 | static int nfs_fsync_dir(struct file *, int); |
f0dd2136 | 59 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
d1bacf9e | 60 | static int nfs_readdir_clear_array(struct page*, gfp_t); |
1da177e4 | 61 | |
4b6f5d20 | 62 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 63 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
64 | .read = generic_read_dir, |
65 | .readdir = nfs_readdir, | |
66 | .open = nfs_opendir, | |
67 | .release = nfs_release, | |
68 | .fsync = nfs_fsync_dir, | |
69 | }; | |
70 | ||
92e1d5be | 71 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
72 | .create = nfs_create, |
73 | .lookup = nfs_lookup, | |
74 | .link = nfs_link, | |
75 | .unlink = nfs_unlink, | |
76 | .symlink = nfs_symlink, | |
77 | .mkdir = nfs_mkdir, | |
78 | .rmdir = nfs_rmdir, | |
79 | .mknod = nfs_mknod, | |
80 | .rename = nfs_rename, | |
81 | .permission = nfs_permission, | |
82 | .getattr = nfs_getattr, | |
83 | .setattr = nfs_setattr, | |
84 | }; | |
85 | ||
d1bacf9e BS |
86 | const struct address_space_operations nfs_dir_addr_space_ops = { |
87 | .releasepage = nfs_readdir_clear_array, | |
88 | }; | |
89 | ||
b7fa0554 | 90 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 91 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
92 | .create = nfs_create, |
93 | .lookup = nfs_lookup, | |
94 | .link = nfs_link, | |
95 | .unlink = nfs_unlink, | |
96 | .symlink = nfs_symlink, | |
97 | .mkdir = nfs_mkdir, | |
98 | .rmdir = nfs_rmdir, | |
99 | .mknod = nfs_mknod, | |
100 | .rename = nfs_rename, | |
101 | .permission = nfs_permission, | |
102 | .getattr = nfs_getattr, | |
103 | .setattr = nfs_setattr, | |
104 | .listxattr = nfs3_listxattr, | |
105 | .getxattr = nfs3_getxattr, | |
106 | .setxattr = nfs3_setxattr, | |
107 | .removexattr = nfs3_removexattr, | |
108 | }; | |
109 | #endif /* CONFIG_NFS_V3 */ | |
110 | ||
1da177e4 LT |
111 | #ifdef CONFIG_NFS_V4 |
112 | ||
113 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
c0204fd2 | 114 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd); |
92e1d5be | 115 | const struct inode_operations nfs4_dir_inode_operations = { |
c0204fd2 | 116 | .create = nfs_open_create, |
1da177e4 LT |
117 | .lookup = nfs_atomic_lookup, |
118 | .link = nfs_link, | |
119 | .unlink = nfs_unlink, | |
120 | .symlink = nfs_symlink, | |
121 | .mkdir = nfs_mkdir, | |
122 | .rmdir = nfs_rmdir, | |
123 | .mknod = nfs_mknod, | |
124 | .rename = nfs_rename, | |
125 | .permission = nfs_permission, | |
126 | .getattr = nfs_getattr, | |
127 | .setattr = nfs_setattr, | |
6b3b5496 BF |
128 | .getxattr = nfs4_getxattr, |
129 | .setxattr = nfs4_setxattr, | |
130 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
131 | }; |
132 | ||
133 | #endif /* CONFIG_NFS_V4 */ | |
134 | ||
135 | /* | |
136 | * Open file | |
137 | */ | |
138 | static int | |
139 | nfs_opendir(struct inode *inode, struct file *filp) | |
140 | { | |
7451c4f0 | 141 | int res; |
1da177e4 | 142 | |
6da24bc9 | 143 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
cc0dd2d1 CL |
144 | filp->f_path.dentry->d_parent->d_name.name, |
145 | filp->f_path.dentry->d_name.name); | |
146 | ||
147 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 148 | |
1da177e4 | 149 | /* Call generic open code in order to cache credentials */ |
7451c4f0 | 150 | res = nfs_open(inode, filp); |
f5a73672 NB |
151 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
152 | /* This is a mountpoint, so d_revalidate will never | |
153 | * have been called, so we need to refresh the | |
154 | * inode (for close-open consistency) ourselves. | |
155 | */ | |
156 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
157 | } | |
1da177e4 LT |
158 | return res; |
159 | } | |
160 | ||
d1bacf9e BS |
161 | struct nfs_cache_array_entry { |
162 | u64 cookie; | |
163 | u64 ino; | |
164 | struct qstr string; | |
165 | }; | |
166 | ||
167 | struct nfs_cache_array { | |
168 | unsigned int size; | |
169 | int eof_index; | |
170 | u64 last_cookie; | |
171 | struct nfs_cache_array_entry array[0]; | |
172 | }; | |
173 | ||
174 | #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry)) | |
175 | ||
82f2e547 | 176 | typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int); |
1da177e4 LT |
177 | typedef struct { |
178 | struct file *file; | |
179 | struct page *page; | |
180 | unsigned long page_index; | |
f0dd2136 TM |
181 | u64 *dir_cookie; |
182 | loff_t current_index; | |
1da177e4 | 183 | decode_dirent_t decode; |
d1bacf9e | 184 | |
1f4eab7e | 185 | unsigned long timestamp; |
4704f0e2 | 186 | unsigned long gencount; |
d1bacf9e BS |
187 | unsigned int cache_entry_index; |
188 | unsigned int plus:1; | |
189 | unsigned int eof:1; | |
1da177e4 LT |
190 | } nfs_readdir_descriptor_t; |
191 | ||
d1bacf9e BS |
192 | /* |
193 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
1da177e4 LT |
194 | */ |
195 | static | |
d1bacf9e BS |
196 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
197 | { | |
8cd51a0c | 198 | void *ptr; |
d1bacf9e BS |
199 | if (page == NULL) |
200 | return ERR_PTR(-EIO); | |
8cd51a0c TM |
201 | ptr = kmap(page); |
202 | if (ptr == NULL) | |
203 | return ERR_PTR(-ENOMEM); | |
204 | return ptr; | |
d1bacf9e BS |
205 | } |
206 | ||
207 | static | |
208 | void nfs_readdir_release_array(struct page *page) | |
209 | { | |
210 | kunmap(page); | |
211 | } | |
212 | ||
213 | /* | |
214 | * we are freeing strings created by nfs_add_to_readdir_array() | |
215 | */ | |
216 | static | |
217 | int nfs_readdir_clear_array(struct page *page, gfp_t mask) | |
218 | { | |
219 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
220 | int i; | |
8cd51a0c TM |
221 | |
222 | if (IS_ERR(array)) | |
223 | return PTR_ERR(array); | |
d1bacf9e BS |
224 | for (i = 0; i < array->size; i++) |
225 | kfree(array->array[i].string.name); | |
226 | nfs_readdir_release_array(page); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | /* | |
231 | * the caller is responsible for freeing qstr.name | |
232 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
233 | * nfs_clear_readdir_array() | |
234 | */ | |
235 | static | |
4a201d6e | 236 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
237 | { |
238 | string->len = len; | |
239 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
240 | if (string->name == NULL) |
241 | return -ENOMEM; | |
04e4bd1c CM |
242 | /* |
243 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
244 | * in a page cache page which kmemleak does not scan. | |
245 | */ | |
246 | kmemleak_not_leak(string->name); | |
4a201d6e TM |
247 | string->hash = full_name_hash(name, len); |
248 | return 0; | |
d1bacf9e BS |
249 | } |
250 | ||
251 | static | |
252 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
253 | { | |
254 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
255 | struct nfs_cache_array_entry *cache_entry; |
256 | int ret; | |
257 | ||
d1bacf9e BS |
258 | if (IS_ERR(array)) |
259 | return PTR_ERR(array); | |
8cd51a0c | 260 | ret = -ENOSPC; |
4a201d6e TM |
261 | if (array->size >= MAX_READDIR_ARRAY) |
262 | goto out; | |
d1bacf9e | 263 | |
4a201d6e TM |
264 | cache_entry = &array->array[array->size]; |
265 | cache_entry->cookie = entry->prev_cookie; | |
266 | cache_entry->ino = entry->ino; | |
267 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); | |
268 | if (ret) | |
269 | goto out; | |
d1bacf9e | 270 | array->last_cookie = entry->cookie; |
8cd51a0c | 271 | array->size++; |
d1bacf9e BS |
272 | if (entry->eof == 1) |
273 | array->eof_index = array->size; | |
4a201d6e | 274 | out: |
d1bacf9e | 275 | nfs_readdir_release_array(page); |
4a201d6e | 276 | return ret; |
d1bacf9e BS |
277 | } |
278 | ||
279 | static | |
280 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
281 | { | |
282 | loff_t diff = desc->file->f_pos - desc->current_index; | |
283 | unsigned int index; | |
284 | ||
285 | if (diff < 0) | |
286 | goto out_eof; | |
287 | if (diff >= array->size) { | |
8cd51a0c | 288 | if (array->eof_index >= 0) |
d1bacf9e BS |
289 | goto out_eof; |
290 | desc->current_index += array->size; | |
291 | return -EAGAIN; | |
292 | } | |
293 | ||
294 | index = (unsigned int)diff; | |
295 | *desc->dir_cookie = array->array[index].cookie; | |
296 | desc->cache_entry_index = index; | |
d1bacf9e BS |
297 | return 0; |
298 | out_eof: | |
299 | desc->eof = 1; | |
300 | return -EBADCOOKIE; | |
301 | } | |
302 | ||
303 | static | |
304 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
305 | { | |
306 | int i; | |
307 | int status = -EAGAIN; | |
308 | ||
309 | for (i = 0; i < array->size; i++) { | |
d1bacf9e BS |
310 | if (array->array[i].cookie == *desc->dir_cookie) { |
311 | desc->cache_entry_index = i; | |
312 | status = 0; | |
8cd51a0c | 313 | goto out; |
d1bacf9e BS |
314 | } |
315 | } | |
8cd51a0c TM |
316 | if (i == array->eof_index) { |
317 | desc->eof = 1; | |
318 | status = -EBADCOOKIE; | |
319 | } | |
320 | out: | |
d1bacf9e BS |
321 | return status; |
322 | } | |
323 | ||
324 | static | |
325 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
326 | { | |
327 | struct nfs_cache_array *array; | |
328 | int status = -EBADCOOKIE; | |
329 | ||
330 | if (desc->dir_cookie == NULL) | |
331 | goto out; | |
332 | ||
333 | array = nfs_readdir_get_array(desc->page); | |
334 | if (IS_ERR(array)) { | |
335 | status = PTR_ERR(array); | |
336 | goto out; | |
337 | } | |
338 | ||
339 | if (*desc->dir_cookie == 0) | |
340 | status = nfs_readdir_search_for_pos(array, desc); | |
341 | else | |
342 | status = nfs_readdir_search_for_cookie(array, desc); | |
343 | ||
344 | nfs_readdir_release_array(desc->page); | |
345 | out: | |
346 | return status; | |
347 | } | |
348 | ||
349 | /* Fill a page with xdr information before transferring to the cache page */ | |
350 | static | |
56e4ebf8 | 351 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 352 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 353 | { |
1da177e4 | 354 | struct rpc_cred *cred = nfs_file_cred(file); |
4704f0e2 | 355 | unsigned long timestamp, gencount; |
1da177e4 LT |
356 | int error; |
357 | ||
1da177e4 LT |
358 | again: |
359 | timestamp = jiffies; | |
4704f0e2 | 360 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 361 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
362 | NFS_SERVER(inode)->dtsize, desc->plus); |
363 | if (error < 0) { | |
364 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
365 | if (error == -ENOTSUPP && desc->plus) { | |
366 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 367 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
368 | desc->plus = 0; |
369 | goto again; | |
370 | } | |
371 | goto error; | |
372 | } | |
1f4eab7e | 373 | desc->timestamp = timestamp; |
4704f0e2 | 374 | desc->gencount = gencount; |
d1bacf9e BS |
375 | error: |
376 | return error; | |
1da177e4 LT |
377 | } |
378 | ||
d1bacf9e BS |
379 | /* Fill in an entry based on the xdr code stored in desc->page */ |
380 | static | |
babddc72 | 381 | int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream) |
1da177e4 | 382 | { |
82f2e547 | 383 | __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus); |
1da177e4 LT |
384 | if (IS_ERR(p)) |
385 | return PTR_ERR(p); | |
1da177e4 | 386 | |
d1bacf9e BS |
387 | entry->fattr->time_start = desc->timestamp; |
388 | entry->fattr->gencount = desc->gencount; | |
389 | return 0; | |
1da177e4 LT |
390 | } |
391 | ||
d39ab9de BS |
392 | static |
393 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
394 | { | |
395 | struct nfs_inode *node; | |
396 | if (dentry->d_inode == NULL) | |
397 | goto different; | |
398 | node = NFS_I(dentry->d_inode); | |
399 | if (node->fh.size != entry->fh->size) | |
400 | goto different; | |
401 | if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 0) | |
402 | goto different; | |
403 | return 1; | |
404 | different: | |
405 | return 0; | |
406 | } | |
407 | ||
408 | static | |
409 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
410 | { | |
4a201d6e TM |
411 | struct qstr filename = { |
412 | .len = entry->len, | |
413 | .name = entry->name, | |
414 | }; | |
415 | struct dentry *dentry; | |
416 | struct dentry *alias; | |
d39ab9de BS |
417 | struct inode *dir = parent->d_inode; |
418 | struct inode *inode; | |
419 | ||
4a201d6e TM |
420 | if (filename.name[0] == '.') { |
421 | if (filename.len == 1) | |
422 | return; | |
423 | if (filename.len == 2 && filename.name[1] == '.') | |
424 | return; | |
425 | } | |
426 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 427 | |
4a201d6e | 428 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
429 | if (dentry != NULL) { |
430 | if (nfs_same_file(dentry, entry)) { | |
431 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
432 | goto out; | |
433 | } else { | |
434 | d_drop(dentry); | |
435 | dput(dentry); | |
436 | } | |
437 | } | |
438 | ||
439 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
440 | if (dentry == NULL) |
441 | return; | |
442 | ||
d39ab9de BS |
443 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
444 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
445 | if (IS_ERR(inode)) | |
446 | goto out; | |
447 | ||
448 | alias = d_materialise_unique(dentry, inode); | |
449 | if (IS_ERR(alias)) | |
450 | goto out; | |
451 | else if (alias) { | |
452 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
453 | dput(alias); | |
454 | } else | |
455 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
456 | ||
457 | out: | |
458 | dput(dentry); | |
d39ab9de BS |
459 | } |
460 | ||
d1bacf9e BS |
461 | /* Perform conversion from xdr to cache array */ |
462 | static | |
8cd51a0c | 463 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
56e4ebf8 | 464 | void *xdr_page, struct page *page, unsigned int buflen) |
1da177e4 | 465 | { |
babddc72 BS |
466 | struct xdr_stream stream; |
467 | struct xdr_buf buf; | |
56e4ebf8 | 468 | __be32 *ptr = xdr_page; |
99424380 | 469 | struct nfs_cache_array *array; |
5c346854 TM |
470 | unsigned int count = 0; |
471 | int status; | |
babddc72 BS |
472 | |
473 | buf.head->iov_base = xdr_page; | |
474 | buf.head->iov_len = buflen; | |
475 | buf.tail->iov_len = 0; | |
476 | buf.page_base = 0; | |
477 | buf.page_len = 0; | |
478 | buf.buflen = buf.head->iov_len; | |
479 | buf.len = buf.head->iov_len; | |
480 | ||
481 | xdr_init_decode(&stream, &buf, ptr); | |
482 | ||
99424380 BS |
483 | |
484 | do { | |
485 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
486 | if (status != 0) { |
487 | if (status == -EAGAIN) | |
488 | status = 0; | |
99424380 | 489 | break; |
8cd51a0c | 490 | } |
99424380 | 491 | |
5c346854 TM |
492 | count++; |
493 | ||
d39ab9de BS |
494 | if (desc->plus == 1) |
495 | nfs_prime_dcache(desc->file->f_path.dentry, entry); | |
8cd51a0c TM |
496 | |
497 | status = nfs_readdir_add_to_array(entry, page); | |
498 | if (status != 0) | |
499 | break; | |
99424380 BS |
500 | } while (!entry->eof); |
501 | ||
5c346854 | 502 | if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) { |
99424380 | 503 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
504 | if (!IS_ERR(array)) { |
505 | array->eof_index = array->size; | |
506 | status = 0; | |
507 | nfs_readdir_release_array(page); | |
5c346854 TM |
508 | } else |
509 | status = PTR_ERR(array); | |
1da177e4 | 510 | } |
8cd51a0c | 511 | return status; |
56e4ebf8 BS |
512 | } |
513 | ||
514 | static | |
515 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
516 | { | |
517 | unsigned int i; | |
518 | for (i = 0; i < npages; i++) | |
519 | put_page(pages[i]); | |
520 | } | |
521 | ||
522 | static | |
523 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
524 | unsigned int npages) | |
525 | { | |
526 | vm_unmap_ram(ptr, npages); | |
527 | nfs_readdir_free_pagearray(pages, npages); | |
528 | } | |
529 | ||
530 | /* | |
531 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
532 | * to nfs_readdir_free_large_page | |
533 | */ | |
534 | static | |
535 | void *nfs_readdir_large_page(struct page **pages, unsigned int npages) | |
536 | { | |
537 | void *ptr; | |
538 | unsigned int i; | |
539 | ||
540 | for (i = 0; i < npages; i++) { | |
541 | struct page *page = alloc_page(GFP_KERNEL); | |
542 | if (page == NULL) | |
543 | goto out_freepages; | |
544 | pages[i] = page; | |
545 | } | |
546 | ||
4a201d6e | 547 | ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL); |
56e4ebf8 BS |
548 | if (!IS_ERR_OR_NULL(ptr)) |
549 | return ptr; | |
550 | out_freepages: | |
551 | nfs_readdir_free_pagearray(pages, i); | |
552 | return NULL; | |
1da177e4 LT |
553 | } |
554 | ||
d1bacf9e BS |
555 | static |
556 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 557 | { |
56e4ebf8 BS |
558 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
559 | void *pages_ptr = NULL; | |
d1bacf9e BS |
560 | struct nfs_entry entry; |
561 | struct file *file = desc->file; | |
562 | struct nfs_cache_array *array; | |
8cd51a0c | 563 | int status = -ENOMEM; |
56e4ebf8 | 564 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
565 | |
566 | entry.prev_cookie = 0; | |
567 | entry.cookie = *desc->dir_cookie; | |
568 | entry.eof = 0; | |
569 | entry.fh = nfs_alloc_fhandle(); | |
570 | entry.fattr = nfs_alloc_fattr(); | |
571 | if (entry.fh == NULL || entry.fattr == NULL) | |
572 | goto out; | |
00a92642 | 573 | |
d1bacf9e | 574 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
575 | if (IS_ERR(array)) { |
576 | status = PTR_ERR(array); | |
577 | goto out; | |
578 | } | |
d1bacf9e BS |
579 | memset(array, 0, sizeof(struct nfs_cache_array)); |
580 | array->eof_index = -1; | |
00a92642 | 581 | |
56e4ebf8 BS |
582 | pages_ptr = nfs_readdir_large_page(pages, array_size); |
583 | if (!pages_ptr) | |
d1bacf9e BS |
584 | goto out_release_array; |
585 | do { | |
ac396128 | 586 | unsigned int pglen; |
56e4ebf8 | 587 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 588 | |
d1bacf9e | 589 | if (status < 0) |
00a92642 | 590 | break; |
ac396128 TM |
591 | pglen = status; |
592 | status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen); | |
8cd51a0c TM |
593 | if (status < 0) { |
594 | if (status == -ENOSPC) | |
595 | status = 0; | |
596 | break; | |
597 | } | |
598 | } while (array->eof_index < 0); | |
d1bacf9e | 599 | |
56e4ebf8 | 600 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
601 | out_release_array: |
602 | nfs_readdir_release_array(page); | |
603 | out: | |
604 | nfs_free_fattr(entry.fattr); | |
605 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
606 | return status; |
607 | } | |
608 | ||
609 | /* | |
d1bacf9e BS |
610 | * Now we cache directories properly, by converting xdr information |
611 | * to an array that can be used for lookups later. This results in | |
612 | * fewer cache pages, since we can store more information on each page. | |
613 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 614 | */ |
d1bacf9e BS |
615 | static |
616 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 617 | { |
01cce933 | 618 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
8cd51a0c | 619 | int ret; |
1da177e4 | 620 | |
8cd51a0c TM |
621 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
622 | if (ret < 0) | |
d1bacf9e BS |
623 | goto error; |
624 | SetPageUptodate(page); | |
1da177e4 | 625 | |
d1bacf9e BS |
626 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
627 | /* Should never happen */ | |
628 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 629 | } |
d1bacf9e BS |
630 | unlock_page(page); |
631 | return 0; | |
632 | error: | |
633 | unlock_page(page); | |
8cd51a0c | 634 | return ret; |
d1bacf9e | 635 | } |
1da177e4 | 636 | |
d1bacf9e BS |
637 | static |
638 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
639 | { | |
640 | page_cache_release(desc->page); | |
641 | desc->page = NULL; | |
642 | } | |
643 | ||
644 | static | |
645 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
646 | { | |
8cd51a0c | 647 | return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, |
d1bacf9e | 648 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
649 | } |
650 | ||
651 | /* | |
d1bacf9e | 652 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 653 | */ |
d1bacf9e BS |
654 | static |
655 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
656 | { | |
657 | int res; | |
658 | ||
659 | desc->page = get_cache_page(desc); | |
660 | if (IS_ERR(desc->page)) | |
661 | return PTR_ERR(desc->page); | |
662 | ||
663 | res = nfs_readdir_search_array(desc); | |
664 | if (res == 0) | |
665 | return 0; | |
666 | cache_page_release(desc); | |
667 | return res; | |
668 | } | |
669 | ||
670 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
671 | static inline |
672 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
673 | { | |
8cd51a0c | 674 | int res; |
d1bacf9e | 675 | |
8cd51a0c TM |
676 | if (desc->page_index == 0) |
677 | desc->current_index = 0; | |
d1bacf9e BS |
678 | while (1) { |
679 | res = find_cache_page(desc); | |
1da177e4 LT |
680 | if (res != -EAGAIN) |
681 | break; | |
d1bacf9e | 682 | desc->page_index++; |
1da177e4 | 683 | } |
1da177e4 LT |
684 | return res; |
685 | } | |
686 | ||
687 | static inline unsigned int dt_type(struct inode *inode) | |
688 | { | |
689 | return (inode->i_mode >> 12) & 15; | |
690 | } | |
691 | ||
1da177e4 LT |
692 | /* |
693 | * Once we've found the start of the dirent within a page: fill 'er up... | |
694 | */ | |
695 | static | |
696 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
697 | filldir_t filldir) | |
698 | { | |
699 | struct file *file = desc->file; | |
d1bacf9e BS |
700 | int i = 0; |
701 | int res = 0; | |
702 | struct nfs_cache_array *array = NULL; | |
703 | unsigned int d_type = DT_UNKNOWN; | |
d1bacf9e BS |
704 | |
705 | array = nfs_readdir_get_array(desc->page); | |
e7c58e97 TM |
706 | if (IS_ERR(array)) { |
707 | res = PTR_ERR(array); | |
708 | goto out; | |
709 | } | |
d1bacf9e BS |
710 | |
711 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 712 | struct nfs_cache_array_entry *ent; |
d1bacf9e | 713 | d_type = DT_UNKNOWN; |
1da177e4 | 714 | |
ece0b423 TM |
715 | ent = &array->array[i]; |
716 | if (filldir(dirent, ent->string.name, ent->string.len, | |
717 | file->f_pos, nfs_compat_user_ino64(ent->ino), d_type) < 0) { | |
718 | desc->eof = 1; | |
1da177e4 | 719 | break; |
ece0b423 | 720 | } |
00a92642 | 721 | file->f_pos++; |
d1bacf9e BS |
722 | desc->cache_entry_index = i; |
723 | if (i < (array->size-1)) | |
724 | *desc->dir_cookie = array->array[i+1].cookie; | |
725 | else | |
726 | *desc->dir_cookie = array->last_cookie; | |
1da177e4 | 727 | } |
8cd51a0c TM |
728 | if (i == array->eof_index) |
729 | desc->eof = 1; | |
d1bacf9e BS |
730 | |
731 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 732 | out: |
d1bacf9e | 733 | cache_page_release(desc); |
1e7cb3dc CL |
734 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
735 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
736 | return res; |
737 | } | |
738 | ||
739 | /* | |
740 | * If we cannot find a cookie in our cache, we suspect that this is | |
741 | * because it points to a deleted file, so we ask the server to return | |
742 | * whatever it thinks is the next entry. We then feed this to filldir. | |
743 | * If all goes well, we should then be able to find our way round the | |
744 | * cache on the next call to readdir_search_pagecache(); | |
745 | * | |
746 | * NOTE: we cannot add the anonymous page to the pagecache because | |
747 | * the data it contains might not be page aligned. Besides, | |
748 | * we should already have a complete representation of the | |
749 | * directory in the page cache by the time we get here. | |
750 | */ | |
751 | static inline | |
752 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
753 | filldir_t filldir) | |
754 | { | |
1da177e4 LT |
755 | struct page *page = NULL; |
756 | int status; | |
d1bacf9e | 757 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 758 | |
1e7cb3dc CL |
759 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
760 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
761 | |
762 | page = alloc_page(GFP_HIGHUSER); | |
763 | if (!page) { | |
764 | status = -ENOMEM; | |
765 | goto out; | |
766 | } | |
d1bacf9e | 767 | |
7a8e1dc3 TM |
768 | desc->page_index = 0; |
769 | desc->page = page; | |
770 | ||
85f8607e TM |
771 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
772 | if (status < 0) | |
1da177e4 LT |
773 | goto out_release; |
774 | ||
775 | status = nfs_do_filldir(desc, dirent, filldir); | |
776 | ||
1da177e4 | 777 | out: |
1e7cb3dc | 778 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 779 | __func__, status); |
1da177e4 LT |
780 | return status; |
781 | out_release: | |
d1bacf9e | 782 | cache_page_release(desc); |
1da177e4 LT |
783 | goto out; |
784 | } | |
785 | ||
00a92642 OG |
786 | /* The file offset position represents the dirent entry number. A |
787 | last cookie cache takes care of the common case of reading the | |
788 | whole directory. | |
1da177e4 LT |
789 | */ |
790 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
791 | { | |
01cce933 | 792 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
793 | struct inode *inode = dentry->d_inode; |
794 | nfs_readdir_descriptor_t my_desc, | |
795 | *desc = &my_desc; | |
aa49b4cf | 796 | int res = -ENOMEM; |
1da177e4 | 797 | |
6da24bc9 | 798 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
799 | dentry->d_parent->d_name.name, dentry->d_name.name, |
800 | (long long)filp->f_pos); | |
91d5b470 CL |
801 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
802 | ||
1da177e4 | 803 | /* |
00a92642 | 804 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 805 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
806 | * to either find the entry with the appropriate number or |
807 | * revalidate the cookie. | |
1da177e4 LT |
808 | */ |
809 | memset(desc, 0, sizeof(*desc)); | |
810 | ||
811 | desc->file = filp; | |
cd3758e3 | 812 | desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; |
1da177e4 LT |
813 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
814 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
815 | ||
565277f6 | 816 | nfs_block_sillyrename(dentry); |
1cda707d | 817 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
818 | if (res < 0) |
819 | goto out; | |
820 | ||
d1bacf9e | 821 | while (desc->eof != 1) { |
1da177e4 | 822 | res = readdir_search_pagecache(desc); |
00a92642 | 823 | |
1da177e4 | 824 | if (res == -EBADCOOKIE) { |
ece0b423 | 825 | res = 0; |
1da177e4 | 826 | /* This means either end of directory */ |
d1bacf9e | 827 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
828 | /* Or that the server has 'lost' a cookie */ |
829 | res = uncached_readdir(desc, dirent, filldir); | |
ece0b423 | 830 | if (res == 0) |
1da177e4 LT |
831 | continue; |
832 | } | |
1da177e4 LT |
833 | break; |
834 | } | |
835 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 836 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 837 | nfs_zap_caches(inode); |
baf57a09 | 838 | desc->page_index = 0; |
1da177e4 | 839 | desc->plus = 0; |
d1bacf9e | 840 | desc->eof = 0; |
1da177e4 LT |
841 | continue; |
842 | } | |
843 | if (res < 0) | |
844 | break; | |
845 | ||
846 | res = nfs_do_filldir(desc, dirent, filldir); | |
ece0b423 | 847 | if (res < 0) |
1da177e4 | 848 | break; |
1da177e4 | 849 | } |
fccca7fc | 850 | out: |
565277f6 | 851 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
852 | if (res > 0) |
853 | res = 0; | |
aa49b4cf | 854 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
855 | dentry->d_parent->d_name.name, dentry->d_name.name, |
856 | res); | |
857 | return res; | |
1da177e4 LT |
858 | } |
859 | ||
10afec90 | 860 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 861 | { |
b84e06c5 CL |
862 | struct dentry *dentry = filp->f_path.dentry; |
863 | struct inode *inode = dentry->d_inode; | |
864 | ||
6da24bc9 | 865 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
866 | dentry->d_parent->d_name.name, |
867 | dentry->d_name.name, | |
868 | offset, origin); | |
869 | ||
870 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
871 | switch (origin) { |
872 | case 1: | |
873 | offset += filp->f_pos; | |
874 | case 0: | |
875 | if (offset >= 0) | |
876 | break; | |
877 | default: | |
878 | offset = -EINVAL; | |
879 | goto out; | |
880 | } | |
881 | if (offset != filp->f_pos) { | |
882 | filp->f_pos = offset; | |
cd3758e3 | 883 | nfs_file_open_context(filp)->dir_cookie = 0; |
f0dd2136 TM |
884 | } |
885 | out: | |
b84e06c5 | 886 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
887 | return offset; |
888 | } | |
889 | ||
1da177e4 LT |
890 | /* |
891 | * All directory operations under NFS are synchronous, so fsync() | |
892 | * is a dummy operation. | |
893 | */ | |
7ea80859 | 894 | static int nfs_fsync_dir(struct file *filp, int datasync) |
1da177e4 | 895 | { |
7ea80859 CH |
896 | struct dentry *dentry = filp->f_path.dentry; |
897 | ||
6da24bc9 | 898 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
899 | dentry->d_parent->d_name.name, dentry->d_name.name, |
900 | datasync); | |
901 | ||
54917786 | 902 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
1da177e4 LT |
903 | return 0; |
904 | } | |
905 | ||
bfc69a45 TM |
906 | /** |
907 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
908 | * @dir - pointer to directory inode | |
909 | * | |
910 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
911 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
912 | * on the server that might have invalidated our dcache. | |
913 | * | |
914 | * The caller should be holding dir->i_lock | |
915 | */ | |
916 | void nfs_force_lookup_revalidate(struct inode *dir) | |
917 | { | |
011935a0 | 918 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
919 | } |
920 | ||
1da177e4 LT |
921 | /* |
922 | * A check for whether or not the parent directory has changed. | |
923 | * In the case it has, we assume that the dentries are untrustworthy | |
924 | * and may need to be looked up again. | |
925 | */ | |
c79ba787 | 926 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
927 | { |
928 | if (IS_ROOT(dentry)) | |
929 | return 1; | |
4eec952e TM |
930 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
931 | return 0; | |
f2c77f4e TM |
932 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
933 | return 0; | |
934 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
935 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
936 | return 0; | |
937 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
938 | return 0; | |
939 | return 1; | |
1da177e4 LT |
940 | } |
941 | ||
1d6757fb TM |
942 | /* |
943 | * Return the intent data that applies to this particular path component | |
944 | * | |
945 | * Note that the current set of intents only apply to the very last | |
946 | * component of the path. | |
947 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
948 | */ | |
949 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
950 | { | |
951 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
952 | return 0; | |
953 | return nd->flags & mask; | |
954 | } | |
955 | ||
a12802ca TM |
956 | /* |
957 | * Use intent information to check whether or not we're going to do | |
958 | * an O_EXCL create using this path component. | |
959 | */ | |
960 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
961 | { | |
962 | if (NFS_PROTO(dir)->version == 2) | |
963 | return 0; | |
3516586a | 964 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
965 | } |
966 | ||
1d6757fb TM |
967 | /* |
968 | * Inode and filehandle revalidation for lookups. | |
969 | * | |
970 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
971 | * or if the intent information indicates that we're about to open this | |
972 | * particular file and the "nocto" mount flag is not set. | |
973 | * | |
974 | */ | |
1da177e4 LT |
975 | static inline |
976 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
977 | { | |
978 | struct nfs_server *server = NFS_SERVER(inode); | |
979 | ||
4e99a1ff TM |
980 | if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) |
981 | return 0; | |
1da177e4 | 982 | if (nd != NULL) { |
1da177e4 | 983 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 984 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
985 | goto out_force; |
986 | /* This is an open(2) */ | |
1d6757fb | 987 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
988 | !(server->flags & NFS_MOUNT_NOCTO) && |
989 | (S_ISREG(inode->i_mode) || | |
990 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 991 | goto out_force; |
4f48af45 | 992 | return 0; |
1da177e4 LT |
993 | } |
994 | return nfs_revalidate_inode(server, inode); | |
995 | out_force: | |
996 | return __nfs_revalidate_inode(server, inode); | |
997 | } | |
998 | ||
999 | /* | |
1000 | * We judge how long we want to trust negative | |
1001 | * dentries by looking at the parent inode mtime. | |
1002 | * | |
1003 | * If parent mtime has changed, we revalidate, else we wait for a | |
1004 | * period corresponding to the parent's attribute cache timeout value. | |
1005 | */ | |
1006 | static inline | |
1007 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1008 | struct nameidata *nd) | |
1009 | { | |
1da177e4 | 1010 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 1011 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 1012 | return 0; |
4eec952e TM |
1013 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1014 | return 1; | |
1da177e4 LT |
1015 | return !nfs_check_verifier(dir, dentry); |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * This is called every time the dcache has a lookup hit, | |
1020 | * and we should check whether we can really trust that | |
1021 | * lookup. | |
1022 | * | |
1023 | * NOTE! The hit can be a negative hit too, don't assume | |
1024 | * we have an inode! | |
1025 | * | |
1026 | * If the parent directory is seen to have changed, we throw out the | |
1027 | * cached dentry and do a new lookup. | |
1028 | */ | |
1029 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
1030 | { | |
1031 | struct inode *dir; | |
1032 | struct inode *inode; | |
1033 | struct dentry *parent; | |
e1fb4d05 TM |
1034 | struct nfs_fh *fhandle = NULL; |
1035 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1036 | int error; |
1da177e4 LT |
1037 | |
1038 | parent = dget_parent(dentry); | |
1da177e4 | 1039 | dir = parent->d_inode; |
91d5b470 | 1040 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1041 | inode = dentry->d_inode; |
1042 | ||
1043 | if (!inode) { | |
1044 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1045 | goto out_bad; | |
1046 | goto out_valid; | |
1047 | } | |
1048 | ||
1049 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1050 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1051 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1052 | dentry->d_name.name); |
1da177e4 LT |
1053 | goto out_bad; |
1054 | } | |
1055 | ||
15860ab1 TM |
1056 | if (nfs_have_delegation(inode, FMODE_READ)) |
1057 | goto out_set_verifier; | |
1058 | ||
1da177e4 | 1059 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1060 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1061 | if (nfs_lookup_verify_inode(inode, nd)) |
1062 | goto out_zap_parent; | |
1063 | goto out_valid; | |
1064 | } | |
1065 | ||
1066 | if (NFS_STALE(inode)) | |
1067 | goto out_bad; | |
1068 | ||
e1fb4d05 TM |
1069 | error = -ENOMEM; |
1070 | fhandle = nfs_alloc_fhandle(); | |
1071 | fattr = nfs_alloc_fattr(); | |
1072 | if (fhandle == NULL || fattr == NULL) | |
1073 | goto out_error; | |
1074 | ||
1075 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1da177e4 LT |
1076 | if (error) |
1077 | goto out_bad; | |
e1fb4d05 | 1078 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1079 | goto out_bad; |
e1fb4d05 | 1080 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1081 | goto out_bad; |
1082 | ||
e1fb4d05 TM |
1083 | nfs_free_fattr(fattr); |
1084 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1085 | out_set_verifier: |
cf8ba45e | 1086 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1087 | out_valid: |
1da177e4 | 1088 | dput(parent); |
1e7cb3dc | 1089 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1090 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1091 | dentry->d_name.name); |
1da177e4 LT |
1092 | return 1; |
1093 | out_zap_parent: | |
1094 | nfs_zap_caches(dir); | |
1095 | out_bad: | |
a1643a92 | 1096 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1097 | if (inode && S_ISDIR(inode->i_mode)) { |
1098 | /* Purge readdir caches. */ | |
1099 | nfs_zap_caches(inode); | |
1100 | /* If we have submounts, don't unhash ! */ | |
1101 | if (have_submounts(dentry)) | |
1102 | goto out_valid; | |
d9e80b7d AV |
1103 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1104 | goto out_valid; | |
1da177e4 LT |
1105 | shrink_dcache_parent(dentry); |
1106 | } | |
1107 | d_drop(dentry); | |
e1fb4d05 TM |
1108 | nfs_free_fattr(fattr); |
1109 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1110 | dput(parent); |
1e7cb3dc | 1111 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1112 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1113 | dentry->d_name.name); |
1da177e4 | 1114 | return 0; |
e1fb4d05 TM |
1115 | out_error: |
1116 | nfs_free_fattr(fattr); | |
1117 | nfs_free_fhandle(fhandle); | |
1118 | dput(parent); | |
1119 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1120 | __func__, dentry->d_parent->d_name.name, | |
1121 | dentry->d_name.name, error); | |
1122 | return error; | |
1da177e4 LT |
1123 | } |
1124 | ||
1125 | /* | |
1126 | * This is called from dput() when d_count is going to 0. | |
1127 | */ | |
1128 | static int nfs_dentry_delete(struct dentry *dentry) | |
1129 | { | |
1130 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1131 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1132 | dentry->d_flags); | |
1133 | ||
77f11192 TM |
1134 | /* Unhash any dentry with a stale inode */ |
1135 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1136 | return 1; | |
1137 | ||
1da177e4 LT |
1138 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1139 | /* Unhash it, so that ->d_iput() would be called */ | |
1140 | return 1; | |
1141 | } | |
1142 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1143 | /* Unhash it, so that ancestors of killed async unlink | |
1144 | * files will be cleaned up during umount */ | |
1145 | return 1; | |
1146 | } | |
1147 | return 0; | |
1148 | ||
1149 | } | |
1150 | ||
1b83d707 TM |
1151 | static void nfs_drop_nlink(struct inode *inode) |
1152 | { | |
1153 | spin_lock(&inode->i_lock); | |
1154 | if (inode->i_nlink > 0) | |
1155 | drop_nlink(inode); | |
1156 | spin_unlock(&inode->i_lock); | |
1157 | } | |
1158 | ||
1da177e4 LT |
1159 | /* |
1160 | * Called when the dentry loses inode. | |
1161 | * We use it to clean up silly-renamed files. | |
1162 | */ | |
1163 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1164 | { | |
83672d39 NB |
1165 | if (S_ISDIR(inode->i_mode)) |
1166 | /* drop any readdir cache as it could easily be old */ | |
1167 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1168 | ||
1da177e4 | 1169 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1170 | drop_nlink(inode); |
e4eff1a6 | 1171 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1172 | } |
1da177e4 LT |
1173 | iput(inode); |
1174 | } | |
1175 | ||
f786aa90 | 1176 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1177 | .d_revalidate = nfs_lookup_revalidate, |
1178 | .d_delete = nfs_dentry_delete, | |
1179 | .d_iput = nfs_dentry_iput, | |
1180 | }; | |
1181 | ||
1da177e4 LT |
1182 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1183 | { | |
1184 | struct dentry *res; | |
565277f6 | 1185 | struct dentry *parent; |
1da177e4 | 1186 | struct inode *inode = NULL; |
e1fb4d05 TM |
1187 | struct nfs_fh *fhandle = NULL; |
1188 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1189 | int error; |
1da177e4 LT |
1190 | |
1191 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1192 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1193 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1194 | |
1195 | res = ERR_PTR(-ENAMETOOLONG); | |
1196 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1197 | goto out; | |
1198 | ||
1da177e4 LT |
1199 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
1200 | ||
fd684071 TM |
1201 | /* |
1202 | * If we're doing an exclusive create, optimize away the lookup | |
1203 | * but don't hash the dentry. | |
1204 | */ | |
1205 | if (nfs_is_exclusive_create(dir, nd)) { | |
1206 | d_instantiate(dentry, NULL); | |
1207 | res = NULL; | |
fc0f684c | 1208 | goto out; |
fd684071 | 1209 | } |
1da177e4 | 1210 | |
e1fb4d05 TM |
1211 | res = ERR_PTR(-ENOMEM); |
1212 | fhandle = nfs_alloc_fhandle(); | |
1213 | fattr = nfs_alloc_fattr(); | |
1214 | if (fhandle == NULL || fattr == NULL) | |
1215 | goto out; | |
1216 | ||
565277f6 TM |
1217 | parent = dentry->d_parent; |
1218 | /* Protect against concurrent sillydeletes */ | |
1219 | nfs_block_sillyrename(parent); | |
e1fb4d05 | 1220 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1221 | if (error == -ENOENT) |
1222 | goto no_entry; | |
1223 | if (error < 0) { | |
1224 | res = ERR_PTR(error); | |
565277f6 | 1225 | goto out_unblock_sillyrename; |
1da177e4 | 1226 | } |
e1fb4d05 | 1227 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1228 | res = (struct dentry *)inode; |
1229 | if (IS_ERR(res)) | |
565277f6 | 1230 | goto out_unblock_sillyrename; |
54ceac45 | 1231 | |
1da177e4 | 1232 | no_entry: |
54ceac45 | 1233 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1234 | if (res != NULL) { |
1235 | if (IS_ERR(res)) | |
565277f6 | 1236 | goto out_unblock_sillyrename; |
1da177e4 | 1237 | dentry = res; |
9eaef27b | 1238 | } |
1da177e4 | 1239 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1240 | out_unblock_sillyrename: |
1241 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1242 | out: |
e1fb4d05 TM |
1243 | nfs_free_fattr(fattr); |
1244 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1245 | return res; |
1246 | } | |
1247 | ||
1248 | #ifdef CONFIG_NFS_V4 | |
1249 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
1250 | ||
f786aa90 | 1251 | const struct dentry_operations nfs4_dentry_operations = { |
1da177e4 LT |
1252 | .d_revalidate = nfs_open_revalidate, |
1253 | .d_delete = nfs_dentry_delete, | |
1254 | .d_iput = nfs_dentry_iput, | |
1255 | }; | |
1256 | ||
1d6757fb TM |
1257 | /* |
1258 | * Use intent information to determine whether we need to substitute | |
1259 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1260 | */ | |
5584c306 | 1261 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1262 | { |
1d6757fb | 1263 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1264 | return 0; |
1265 | /* NFS does not (yet) have a stateful open for directories */ | |
1266 | if (nd->flags & LOOKUP_DIRECTORY) | |
1267 | return 0; | |
1268 | /* Are we trying to write to a read only partition? */ | |
2c463e95 DH |
1269 | if (__mnt_is_readonly(nd->path.mnt) && |
1270 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
1da177e4 LT |
1271 | return 0; |
1272 | return 1; | |
1273 | } | |
1274 | ||
cd9a1c0e TM |
1275 | static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd) |
1276 | { | |
1277 | struct path path = { | |
1278 | .mnt = nd->path.mnt, | |
1279 | .dentry = dentry, | |
1280 | }; | |
1281 | struct nfs_open_context *ctx; | |
1282 | struct rpc_cred *cred; | |
1283 | fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); | |
1284 | ||
1285 | cred = rpc_lookup_cred(); | |
1286 | if (IS_ERR(cred)) | |
1287 | return ERR_CAST(cred); | |
1288 | ctx = alloc_nfs_open_context(&path, cred, fmode); | |
1289 | put_rpccred(cred); | |
1290 | if (ctx == NULL) | |
1291 | return ERR_PTR(-ENOMEM); | |
1292 | return ctx; | |
1293 | } | |
1294 | ||
1295 | static int do_open(struct inode *inode, struct file *filp) | |
1296 | { | |
1297 | nfs_fscache_set_inode_cookie(inode, filp); | |
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx) | |
1302 | { | |
1303 | struct file *filp; | |
1304 | int ret = 0; | |
1305 | ||
1306 | /* If the open_intent is for execute, we have an extra check to make */ | |
1307 | if (ctx->mode & FMODE_EXEC) { | |
1308 | ret = nfs_may_open(ctx->path.dentry->d_inode, | |
1309 | ctx->cred, | |
1310 | nd->intent.open.flags); | |
1311 | if (ret < 0) | |
1312 | goto out; | |
1313 | } | |
1314 | filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open); | |
1315 | if (IS_ERR(filp)) | |
1316 | ret = PTR_ERR(filp); | |
1317 | else | |
1318 | nfs_file_set_open_context(filp, ctx); | |
1319 | out: | |
1320 | put_nfs_open_context(ctx); | |
1321 | return ret; | |
1322 | } | |
1323 | ||
1da177e4 LT |
1324 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
1325 | { | |
cd9a1c0e TM |
1326 | struct nfs_open_context *ctx; |
1327 | struct iattr attr; | |
1da177e4 | 1328 | struct dentry *res = NULL; |
f46e0bd3 | 1329 | struct inode *inode; |
cd9a1c0e | 1330 | int open_flags; |
898f635c | 1331 | int err; |
1da177e4 | 1332 | |
1e7cb3dc CL |
1333 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
1334 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1335 | ||
1da177e4 | 1336 | /* Check that we are indeed trying to open this file */ |
5584c306 | 1337 | if (!is_atomic_open(nd)) |
1da177e4 LT |
1338 | goto no_open; |
1339 | ||
1340 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1341 | res = ERR_PTR(-ENAMETOOLONG); | |
1342 | goto out; | |
1343 | } | |
1344 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1345 | ||
d4d9cdcb TM |
1346 | /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash |
1347 | * the dentry. */ | |
3516586a | 1348 | if (nd->flags & LOOKUP_EXCL) { |
d4d9cdcb | 1349 | d_instantiate(dentry, NULL); |
02a913a7 TM |
1350 | goto out; |
1351 | } | |
1da177e4 | 1352 | |
cd9a1c0e TM |
1353 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1354 | res = ERR_CAST(ctx); | |
1355 | if (IS_ERR(ctx)) | |
1356 | goto out; | |
1357 | ||
1358 | open_flags = nd->intent.open.flags; | |
1359 | if (nd->flags & LOOKUP_CREATE) { | |
1360 | attr.ia_mode = nd->intent.open.create_mode; | |
1361 | attr.ia_valid = ATTR_MODE; | |
1362 | if (!IS_POSIXACL(dir)) | |
1363 | attr.ia_mode &= ~current_umask(); | |
1364 | } else { | |
898f635c | 1365 | open_flags &= ~(O_EXCL | O_CREAT); |
cd9a1c0e | 1366 | attr.ia_valid = 0; |
cd9a1c0e TM |
1367 | } |
1368 | ||
1da177e4 | 1369 | /* Open the file on the server */ |
f46e0bd3 | 1370 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1371 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
f46e0bd3 TM |
1372 | if (IS_ERR(inode)) { |
1373 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1374 | put_nfs_open_context(ctx); |
f46e0bd3 | 1375 | switch (PTR_ERR(inode)) { |
1da177e4 LT |
1376 | /* Make a negative dentry */ |
1377 | case -ENOENT: | |
f46e0bd3 | 1378 | d_add(dentry, NULL); |
02a913a7 TM |
1379 | res = NULL; |
1380 | goto out; | |
1da177e4 | 1381 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1382 | case -ENOTDIR: |
1383 | goto no_open; | |
1da177e4 LT |
1384 | case -ELOOP: |
1385 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1386 | goto no_open; | |
23ebbd9a | 1387 | /* case -EISDIR: */ |
1da177e4 LT |
1388 | /* case -EINVAL: */ |
1389 | default: | |
f46e0bd3 | 1390 | res = ERR_CAST(inode); |
1da177e4 LT |
1391 | goto out; |
1392 | } | |
cd9a1c0e | 1393 | } |
f46e0bd3 | 1394 | res = d_add_unique(dentry, inode); |
898f635c | 1395 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 TM |
1396 | if (res != NULL) { |
1397 | dput(ctx->path.dentry); | |
1398 | ctx->path.dentry = dget(res); | |
1da177e4 | 1399 | dentry = res; |
f46e0bd3 | 1400 | } |
898f635c TM |
1401 | err = nfs_intent_set_file(nd, ctx); |
1402 | if (err < 0) { | |
1403 | if (res != NULL) | |
1404 | dput(res); | |
1405 | return ERR_PTR(err); | |
1406 | } | |
1da177e4 | 1407 | out: |
f46e0bd3 | 1408 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1409 | return res; |
1410 | no_open: | |
1411 | return nfs_lookup(dir, dentry, nd); | |
1412 | } | |
1413 | ||
1414 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1415 | { | |
1416 | struct dentry *parent = NULL; | |
1417 | struct inode *inode = dentry->d_inode; | |
1418 | struct inode *dir; | |
b8d4cadd | 1419 | struct nfs_open_context *ctx; |
1da177e4 LT |
1420 | int openflags, ret = 0; |
1421 | ||
1f063d2c | 1422 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1423 | goto no_open; |
2b484297 | 1424 | |
1da177e4 LT |
1425 | parent = dget_parent(dentry); |
1426 | dir = parent->d_inode; | |
2b484297 | 1427 | |
1da177e4 LT |
1428 | /* We can't create new files in nfs_open_revalidate(), so we |
1429 | * optimize away revalidation of negative dentries. | |
1430 | */ | |
216d5d06 TM |
1431 | if (inode == NULL) { |
1432 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1433 | ret = 1; | |
1da177e4 | 1434 | goto out; |
216d5d06 TM |
1435 | } |
1436 | ||
1da177e4 LT |
1437 | /* NFS only supports OPEN on regular files */ |
1438 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1439 | goto no_open_dput; |
1da177e4 LT |
1440 | openflags = nd->intent.open.flags; |
1441 | /* We cannot do exclusive creation on a positive dentry */ | |
1442 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1443 | goto no_open_dput; |
1da177e4 | 1444 | /* We can't create new files, or truncate existing ones here */ |
0a377cff | 1445 | openflags &= ~(O_CREAT|O_EXCL|O_TRUNC); |
1da177e4 | 1446 | |
b8d4cadd TM |
1447 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1448 | ret = PTR_ERR(ctx); | |
1449 | if (IS_ERR(ctx)) | |
1450 | goto out; | |
1da177e4 | 1451 | /* |
1b1dcc1b | 1452 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1453 | * operations that change the directory. We therefore save the |
1454 | * change attribute *before* we do the RPC call. | |
1455 | */ | |
2b484297 | 1456 | inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL); |
535918f1 TM |
1457 | if (IS_ERR(inode)) { |
1458 | ret = PTR_ERR(inode); | |
1459 | switch (ret) { | |
1460 | case -EPERM: | |
1461 | case -EACCES: | |
1462 | case -EDQUOT: | |
1463 | case -ENOSPC: | |
1464 | case -EROFS: | |
1465 | goto out_put_ctx; | |
1466 | default: | |
1467 | goto out_drop; | |
1468 | } | |
1469 | } | |
1470 | iput(inode); | |
898f635c | 1471 | if (inode != dentry->d_inode) |
535918f1 | 1472 | goto out_drop; |
898f635c TM |
1473 | |
1474 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1475 | ret = nfs_intent_set_file(nd, ctx); | |
1476 | if (ret >= 0) | |
1477 | ret = 1; | |
1da177e4 LT |
1478 | out: |
1479 | dput(parent); | |
1da177e4 | 1480 | return ret; |
535918f1 TM |
1481 | out_drop: |
1482 | d_drop(dentry); | |
1483 | ret = 0; | |
1484 | out_put_ctx: | |
1485 | put_nfs_open_context(ctx); | |
1486 | goto out; | |
1487 | ||
5584c306 | 1488 | no_open_dput: |
1da177e4 | 1489 | dput(parent); |
5584c306 | 1490 | no_open: |
1da177e4 LT |
1491 | return nfs_lookup_revalidate(dentry, nd); |
1492 | } | |
c0204fd2 TM |
1493 | |
1494 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, | |
1495 | struct nameidata *nd) | |
1496 | { | |
1497 | struct nfs_open_context *ctx = NULL; | |
1498 | struct iattr attr; | |
1499 | int error; | |
1500 | int open_flags = 0; | |
1501 | ||
1502 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1503 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1504 | ||
1505 | attr.ia_mode = mode; | |
1506 | attr.ia_valid = ATTR_MODE; | |
1507 | ||
1508 | if ((nd->flags & LOOKUP_CREATE) != 0) { | |
1509 | open_flags = nd->intent.open.flags; | |
1510 | ||
1511 | ctx = nameidata_to_nfs_open_context(dentry, nd); | |
1512 | error = PTR_ERR(ctx); | |
1513 | if (IS_ERR(ctx)) | |
898f635c | 1514 | goto out_err_drop; |
c0204fd2 TM |
1515 | } |
1516 | ||
1517 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1518 | if (error != 0) | |
1519 | goto out_put_ctx; | |
898f635c TM |
1520 | if (ctx != NULL) { |
1521 | error = nfs_intent_set_file(nd, ctx); | |
1522 | if (error < 0) | |
1523 | goto out_err; | |
1524 | } | |
c0204fd2 TM |
1525 | return 0; |
1526 | out_put_ctx: | |
1527 | if (ctx != NULL) | |
1528 | put_nfs_open_context(ctx); | |
898f635c | 1529 | out_err_drop: |
c0204fd2 | 1530 | d_drop(dentry); |
898f635c | 1531 | out_err: |
c0204fd2 TM |
1532 | return error; |
1533 | } | |
1534 | ||
1da177e4 LT |
1535 | #endif /* CONFIG_NFSV4 */ |
1536 | ||
1da177e4 LT |
1537 | /* |
1538 | * Code common to create, mkdir, and mknod. | |
1539 | */ | |
1540 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1541 | struct nfs_fattr *fattr) | |
1542 | { | |
fab728e1 TM |
1543 | struct dentry *parent = dget_parent(dentry); |
1544 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1545 | struct inode *inode; |
1546 | int error = -EACCES; | |
1547 | ||
fab728e1 TM |
1548 | d_drop(dentry); |
1549 | ||
1da177e4 LT |
1550 | /* We may have been initialized further down */ |
1551 | if (dentry->d_inode) | |
fab728e1 | 1552 | goto out; |
1da177e4 | 1553 | if (fhandle->size == 0) { |
1da177e4 LT |
1554 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1555 | if (error) | |
fab728e1 | 1556 | goto out_error; |
1da177e4 | 1557 | } |
5724ab37 | 1558 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1559 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1560 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1561 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1562 | if (error < 0) |
fab728e1 | 1563 | goto out_error; |
1da177e4 | 1564 | } |
1da177e4 | 1565 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1566 | error = PTR_ERR(inode); |
1567 | if (IS_ERR(inode)) | |
fab728e1 TM |
1568 | goto out_error; |
1569 | d_add(dentry, inode); | |
1570 | out: | |
1571 | dput(parent); | |
1da177e4 | 1572 | return 0; |
fab728e1 TM |
1573 | out_error: |
1574 | nfs_mark_for_revalidate(dir); | |
1575 | dput(parent); | |
1576 | return error; | |
1da177e4 LT |
1577 | } |
1578 | ||
1579 | /* | |
1580 | * Following a failed create operation, we drop the dentry rather | |
1581 | * than retain a negative dentry. This avoids a problem in the event | |
1582 | * that the operation succeeded on the server, but an error in the | |
1583 | * reply path made it appear to have failed. | |
1584 | */ | |
1585 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1586 | struct nameidata *nd) | |
1587 | { | |
1588 | struct iattr attr; | |
1589 | int error; | |
1da177e4 | 1590 | |
1e7cb3dc CL |
1591 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1592 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1593 | |
1594 | attr.ia_mode = mode; | |
1595 | attr.ia_valid = ATTR_MODE; | |
1596 | ||
c0204fd2 | 1597 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL); |
1da177e4 LT |
1598 | if (error != 0) |
1599 | goto out_err; | |
1da177e4 LT |
1600 | return 0; |
1601 | out_err: | |
1da177e4 LT |
1602 | d_drop(dentry); |
1603 | return error; | |
1604 | } | |
1605 | ||
1606 | /* | |
1607 | * See comments for nfs_proc_create regarding failed operations. | |
1608 | */ | |
1609 | static int | |
1610 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1611 | { | |
1612 | struct iattr attr; | |
1613 | int status; | |
1614 | ||
1e7cb3dc CL |
1615 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1616 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1617 | |
1618 | if (!new_valid_dev(rdev)) | |
1619 | return -EINVAL; | |
1620 | ||
1621 | attr.ia_mode = mode; | |
1622 | attr.ia_valid = ATTR_MODE; | |
1623 | ||
1da177e4 | 1624 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1625 | if (status != 0) |
1626 | goto out_err; | |
1da177e4 LT |
1627 | return 0; |
1628 | out_err: | |
1da177e4 LT |
1629 | d_drop(dentry); |
1630 | return status; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * See comments for nfs_proc_create regarding failed operations. | |
1635 | */ | |
1636 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1637 | { | |
1638 | struct iattr attr; | |
1639 | int error; | |
1640 | ||
1e7cb3dc CL |
1641 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1642 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1643 | |
1644 | attr.ia_valid = ATTR_MODE; | |
1645 | attr.ia_mode = mode | S_IFDIR; | |
1646 | ||
1da177e4 | 1647 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1648 | if (error != 0) |
1649 | goto out_err; | |
1da177e4 LT |
1650 | return 0; |
1651 | out_err: | |
1652 | d_drop(dentry); | |
1da177e4 LT |
1653 | return error; |
1654 | } | |
1655 | ||
d45b9d8b TM |
1656 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1657 | { | |
1658 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1659 | d_delete(dentry); | |
1660 | } | |
1661 | ||
1da177e4 LT |
1662 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1663 | { | |
1664 | int error; | |
1665 | ||
1e7cb3dc CL |
1666 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1667 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1668 | |
1da177e4 LT |
1669 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1670 | /* Ensure the VFS deletes this inode */ | |
1671 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1672 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1673 | else if (error == -ENOENT) |
1674 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1675 | |
1676 | return error; | |
1677 | } | |
1678 | ||
1da177e4 LT |
1679 | /* |
1680 | * Remove a file after making sure there are no pending writes, | |
1681 | * and after checking that the file has only one user. | |
1682 | * | |
1683 | * We invalidate the attribute cache and free the inode prior to the operation | |
1684 | * to avoid possible races if the server reuses the inode. | |
1685 | */ | |
1686 | static int nfs_safe_remove(struct dentry *dentry) | |
1687 | { | |
1688 | struct inode *dir = dentry->d_parent->d_inode; | |
1689 | struct inode *inode = dentry->d_inode; | |
1690 | int error = -EBUSY; | |
1691 | ||
1692 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1693 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1694 | ||
1695 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1696 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1697 | error = 0; | |
1698 | goto out; | |
1699 | } | |
1700 | ||
1da177e4 | 1701 | if (inode != NULL) { |
cae7a073 | 1702 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1703 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1704 | /* The VFS may want to delete this inode */ | |
1705 | if (error == 0) | |
1b83d707 | 1706 | nfs_drop_nlink(inode); |
5ba7cc48 | 1707 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1708 | } else |
1709 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1710 | if (error == -ENOENT) |
1711 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1712 | out: |
1713 | return error; | |
1714 | } | |
1715 | ||
1716 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1717 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1718 | * | |
1719 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1720 | */ | |
1721 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1722 | { | |
1723 | int error; | |
1724 | int need_rehash = 0; | |
1725 | ||
1726 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1727 | dir->i_ino, dentry->d_name.name); | |
1728 | ||
1da177e4 LT |
1729 | spin_lock(&dcache_lock); |
1730 | spin_lock(&dentry->d_lock); | |
1731 | if (atomic_read(&dentry->d_count) > 1) { | |
1732 | spin_unlock(&dentry->d_lock); | |
1733 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1734 | /* Start asynchronous writeout of the inode */ |
1735 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1736 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1737 | return error; |
1738 | } | |
1739 | if (!d_unhashed(dentry)) { | |
1740 | __d_drop(dentry); | |
1741 | need_rehash = 1; | |
1742 | } | |
1743 | spin_unlock(&dentry->d_lock); | |
1744 | spin_unlock(&dcache_lock); | |
1745 | error = nfs_safe_remove(dentry); | |
d45b9d8b | 1746 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1747 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1748 | } else if (need_rehash) | |
1749 | d_rehash(dentry); | |
1da177e4 LT |
1750 | return error; |
1751 | } | |
1752 | ||
873101b3 CL |
1753 | /* |
1754 | * To create a symbolic link, most file systems instantiate a new inode, | |
1755 | * add a page to it containing the path, then write it out to the disk | |
1756 | * using prepare_write/commit_write. | |
1757 | * | |
1758 | * Unfortunately the NFS client can't create the in-core inode first | |
1759 | * because it needs a file handle to create an in-core inode (see | |
1760 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1761 | * symlink request has completed on the server. | |
1762 | * | |
1763 | * So instead we allocate a raw page, copy the symname into it, then do | |
1764 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1765 | * now have a new file handle and can instantiate an in-core NFS inode | |
1766 | * and move the raw page into its mapping. | |
1767 | */ | |
1768 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1769 | { |
873101b3 CL |
1770 | struct pagevec lru_pvec; |
1771 | struct page *page; | |
1772 | char *kaddr; | |
1da177e4 | 1773 | struct iattr attr; |
873101b3 | 1774 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1775 | int error; |
1776 | ||
1777 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1778 | dir->i_ino, dentry->d_name.name, symname); | |
1779 | ||
873101b3 CL |
1780 | if (pathlen > PAGE_SIZE) |
1781 | return -ENAMETOOLONG; | |
1da177e4 | 1782 | |
873101b3 CL |
1783 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1784 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1785 | |
83d93f22 | 1786 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1787 | if (!page) |
873101b3 | 1788 | return -ENOMEM; |
873101b3 CL |
1789 | |
1790 | kaddr = kmap_atomic(page, KM_USER0); | |
1791 | memcpy(kaddr, symname, pathlen); | |
1792 | if (pathlen < PAGE_SIZE) | |
1793 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1794 | kunmap_atomic(kaddr, KM_USER0); | |
1795 | ||
94a6d753 | 1796 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1797 | if (error != 0) { |
1798 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1799 | dir->i_sb->s_id, dir->i_ino, | |
1800 | dentry->d_name.name, symname, error); | |
1da177e4 | 1801 | d_drop(dentry); |
873101b3 | 1802 | __free_page(page); |
873101b3 CL |
1803 | return error; |
1804 | } | |
1805 | ||
1806 | /* | |
1807 | * No big deal if we can't add this page to the page cache here. | |
1808 | * READLINK will get the missing page from the server if needed. | |
1809 | */ | |
1810 | pagevec_init(&lru_pvec, 0); | |
1811 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1812 | GFP_KERNEL)) { | |
39cf8a13 | 1813 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1814 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1815 | SetPageUptodate(page); |
1816 | unlock_page(page); | |
1817 | } else | |
1818 | __free_page(page); | |
1819 | ||
873101b3 | 1820 | return 0; |
1da177e4 LT |
1821 | } |
1822 | ||
1823 | static int | |
1824 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1825 | { | |
1826 | struct inode *inode = old_dentry->d_inode; | |
1827 | int error; | |
1828 | ||
1829 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1830 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1831 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1832 | ||
9a3936aa TM |
1833 | nfs_inode_return_delegation(inode); |
1834 | ||
9697d234 | 1835 | d_drop(dentry); |
1da177e4 | 1836 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1837 | if (error == 0) { |
7de9c6ee | 1838 | ihold(inode); |
9697d234 | 1839 | d_add(dentry, inode); |
cf809556 | 1840 | } |
1da177e4 LT |
1841 | return error; |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * RENAME | |
1846 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1847 | * different file handle for the same inode after a rename (e.g. when | |
1848 | * moving to a different directory). A fail-safe method to do so would | |
1849 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1850 | * rename the old file using the sillyrename stuff. This way, the original | |
1851 | * file in old_dir will go away when the last process iput()s the inode. | |
1852 | * | |
1853 | * FIXED. | |
1854 | * | |
1855 | * It actually works quite well. One needs to have the possibility for | |
1856 | * at least one ".nfs..." file in each directory the file ever gets | |
1857 | * moved or linked to which happens automagically with the new | |
1858 | * implementation that only depends on the dcache stuff instead of | |
1859 | * using the inode layer | |
1860 | * | |
1861 | * Unfortunately, things are a little more complicated than indicated | |
1862 | * above. For a cross-directory move, we want to make sure we can get | |
1863 | * rid of the old inode after the operation. This means there must be | |
1864 | * no pending writes (if it's a file), and the use count must be 1. | |
1865 | * If these conditions are met, we can drop the dentries before doing | |
1866 | * the rename. | |
1867 | */ | |
1868 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1869 | struct inode *new_dir, struct dentry *new_dentry) | |
1870 | { | |
1871 | struct inode *old_inode = old_dentry->d_inode; | |
1872 | struct inode *new_inode = new_dentry->d_inode; | |
1873 | struct dentry *dentry = NULL, *rehash = NULL; | |
1874 | int error = -EBUSY; | |
1875 | ||
1da177e4 LT |
1876 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1877 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1878 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1879 | atomic_read(&new_dentry->d_count)); | |
1880 | ||
1881 | /* | |
28f79a1a MS |
1882 | * For non-directories, check whether the target is busy and if so, |
1883 | * make a copy of the dentry and then do a silly-rename. If the | |
1884 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1885 | * the new target. | |
1da177e4 | 1886 | */ |
27226104 MS |
1887 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1888 | /* | |
1889 | * To prevent any new references to the target during the | |
1890 | * rename, we unhash the dentry in advance. | |
1891 | */ | |
1892 | if (!d_unhashed(new_dentry)) { | |
1893 | d_drop(new_dentry); | |
1894 | rehash = new_dentry; | |
1895 | } | |
1da177e4 | 1896 | |
27226104 MS |
1897 | if (atomic_read(&new_dentry->d_count) > 2) { |
1898 | int err; | |
1899 | ||
1900 | /* copy the target dentry's name */ | |
1901 | dentry = d_alloc(new_dentry->d_parent, | |
1902 | &new_dentry->d_name); | |
1903 | if (!dentry) | |
1904 | goto out; | |
1905 | ||
1906 | /* silly-rename the existing target ... */ | |
1907 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1908 | if (err) |
27226104 | 1909 | goto out; |
24e93025 MS |
1910 | |
1911 | new_dentry = dentry; | |
56335936 | 1912 | rehash = NULL; |
24e93025 | 1913 | new_inode = NULL; |
27226104 | 1914 | } |
b1e4adf4 | 1915 | } |
1da177e4 | 1916 | |
cae7a073 | 1917 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 1918 | if (new_inode != NULL) |
24174119 | 1919 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 1920 | |
1da177e4 LT |
1921 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1922 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1923 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1924 | out: |
1925 | if (rehash) | |
1926 | d_rehash(rehash); | |
1927 | if (!error) { | |
b1e4adf4 TM |
1928 | if (new_inode != NULL) |
1929 | nfs_drop_nlink(new_inode); | |
349457cc | 1930 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1931 | nfs_set_verifier(new_dentry, |
1932 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1933 | } else if (error == -ENOENT) |
1934 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1935 | |
1936 | /* new dentry created? */ | |
1937 | if (dentry) | |
1938 | dput(dentry); | |
1da177e4 LT |
1939 | return error; |
1940 | } | |
1941 | ||
cfcea3e8 TM |
1942 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1943 | static LIST_HEAD(nfs_access_lru_list); | |
1944 | static atomic_long_t nfs_access_nr_entries; | |
1945 | ||
1c3c07e9 TM |
1946 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1947 | { | |
1948 | put_rpccred(entry->cred); | |
1949 | kfree(entry); | |
cfcea3e8 TM |
1950 | smp_mb__before_atomic_dec(); |
1951 | atomic_long_dec(&nfs_access_nr_entries); | |
1952 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1953 | } |
1954 | ||
1a81bb8a TM |
1955 | static void nfs_access_free_list(struct list_head *head) |
1956 | { | |
1957 | struct nfs_access_entry *cache; | |
1958 | ||
1959 | while (!list_empty(head)) { | |
1960 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1961 | list_del(&cache->lru); | |
1962 | nfs_access_free_entry(cache); | |
1963 | } | |
1964 | } | |
1965 | ||
7f8275d0 | 1966 | int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
979df72e TM |
1967 | { |
1968 | LIST_HEAD(head); | |
aa510da5 | 1969 | struct nfs_inode *nfsi, *next; |
979df72e TM |
1970 | struct nfs_access_entry *cache; |
1971 | ||
61d5eb29 TM |
1972 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1973 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 1974 | |
a50f7951 | 1975 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 1976 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
1977 | struct inode *inode; |
1978 | ||
1979 | if (nr_to_scan-- == 0) | |
1980 | break; | |
9c7e7e23 | 1981 | inode = &nfsi->vfs_inode; |
979df72e TM |
1982 | spin_lock(&inode->i_lock); |
1983 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1984 | goto remove_lru_entry; | |
1985 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1986 | struct nfs_access_entry, lru); | |
1987 | list_move(&cache->lru, &head); | |
1988 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1989 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1990 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1991 | &nfs_access_lru_list); | |
1992 | else { | |
1993 | remove_lru_entry: | |
1994 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 1995 | smp_mb__before_clear_bit(); |
979df72e | 1996 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 1997 | smp_mb__after_clear_bit(); |
979df72e | 1998 | } |
59844a9b | 1999 | spin_unlock(&inode->i_lock); |
979df72e TM |
2000 | } |
2001 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2002 | nfs_access_free_list(&head); |
979df72e TM |
2003 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
2004 | } | |
2005 | ||
1a81bb8a | 2006 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2007 | { |
1c3c07e9 | 2008 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2009 | struct rb_node *n; |
1c3c07e9 TM |
2010 | struct nfs_access_entry *entry; |
2011 | ||
2012 | /* Unhook entries from the cache */ | |
2013 | while ((n = rb_first(root_node)) != NULL) { | |
2014 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2015 | rb_erase(n, root_node); | |
1a81bb8a | 2016 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2017 | } |
2018 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2019 | } |
2020 | ||
1c3c07e9 | 2021 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2022 | { |
1a81bb8a TM |
2023 | LIST_HEAD(head); |
2024 | ||
2025 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2026 | return; | |
cfcea3e8 | 2027 | /* Remove from global LRU init */ |
1a81bb8a TM |
2028 | spin_lock(&nfs_access_lru_lock); |
2029 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2030 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2031 | |
1c3c07e9 | 2032 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2033 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2034 | spin_unlock(&inode->i_lock); | |
2035 | spin_unlock(&nfs_access_lru_lock); | |
2036 | nfs_access_free_list(&head); | |
1c3c07e9 | 2037 | } |
1da177e4 | 2038 | |
1c3c07e9 TM |
2039 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2040 | { | |
2041 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2042 | struct nfs_access_entry *entry; | |
2043 | ||
2044 | while (n != NULL) { | |
2045 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2046 | ||
2047 | if (cred < entry->cred) | |
2048 | n = n->rb_left; | |
2049 | else if (cred > entry->cred) | |
2050 | n = n->rb_right; | |
2051 | else | |
2052 | return entry; | |
1da177e4 | 2053 | } |
1c3c07e9 TM |
2054 | return NULL; |
2055 | } | |
2056 | ||
af22f94a | 2057 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2058 | { |
2059 | struct nfs_inode *nfsi = NFS_I(inode); | |
2060 | struct nfs_access_entry *cache; | |
2061 | int err = -ENOENT; | |
2062 | ||
dc59250c | 2063 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2064 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2065 | goto out_zap; | |
2066 | cache = nfs_access_search_rbtree(inode, cred); | |
2067 | if (cache == NULL) | |
2068 | goto out; | |
b4d2314b | 2069 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2070 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2071 | goto out_stale; |
2072 | res->jiffies = cache->jiffies; | |
2073 | res->cred = cache->cred; | |
2074 | res->mask = cache->mask; | |
cfcea3e8 | 2075 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2076 | err = 0; |
2077 | out: | |
2078 | spin_unlock(&inode->i_lock); | |
2079 | return err; | |
2080 | out_stale: | |
2081 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2082 | list_del(&cache->lru); |
1c3c07e9 TM |
2083 | spin_unlock(&inode->i_lock); |
2084 | nfs_access_free_entry(cache); | |
2085 | return -ENOENT; | |
2086 | out_zap: | |
1a81bb8a TM |
2087 | spin_unlock(&inode->i_lock); |
2088 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2089 | return -ENOENT; |
2090 | } | |
2091 | ||
2092 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2093 | { | |
cfcea3e8 TM |
2094 | struct nfs_inode *nfsi = NFS_I(inode); |
2095 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2096 | struct rb_node **p = &root_node->rb_node; |
2097 | struct rb_node *parent = NULL; | |
2098 | struct nfs_access_entry *entry; | |
2099 | ||
2100 | spin_lock(&inode->i_lock); | |
2101 | while (*p != NULL) { | |
2102 | parent = *p; | |
2103 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2104 | ||
2105 | if (set->cred < entry->cred) | |
2106 | p = &parent->rb_left; | |
2107 | else if (set->cred > entry->cred) | |
2108 | p = &parent->rb_right; | |
2109 | else | |
2110 | goto found; | |
2111 | } | |
2112 | rb_link_node(&set->rb_node, parent, p); | |
2113 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2114 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2115 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2116 | return; |
2117 | found: | |
2118 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2119 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2120 | list_del(&entry->lru); | |
1c3c07e9 TM |
2121 | spin_unlock(&inode->i_lock); |
2122 | nfs_access_free_entry(entry); | |
2123 | } | |
2124 | ||
af22f94a | 2125 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2126 | { |
2127 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2128 | if (cache == NULL) | |
2129 | return; | |
2130 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2131 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2132 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2133 | cache->mask = set->mask; |
1c3c07e9 TM |
2134 | |
2135 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2136 | |
2137 | /* Update accounting */ | |
2138 | smp_mb__before_atomic_inc(); | |
2139 | atomic_long_inc(&nfs_access_nr_entries); | |
2140 | smp_mb__after_atomic_inc(); | |
2141 | ||
2142 | /* Add inode to global LRU list */ | |
1a81bb8a | 2143 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2144 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2145 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2146 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2147 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2148 | spin_unlock(&nfs_access_lru_lock); |
2149 | } | |
1da177e4 LT |
2150 | } |
2151 | ||
2152 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2153 | { | |
2154 | struct nfs_access_entry cache; | |
2155 | int status; | |
2156 | ||
2157 | status = nfs_access_get_cached(inode, cred, &cache); | |
2158 | if (status == 0) | |
2159 | goto out; | |
2160 | ||
2161 | /* Be clever: ask server to check for all possible rights */ | |
2162 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2163 | cache.cred = cred; | |
2164 | cache.jiffies = jiffies; | |
2165 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2166 | if (status != 0) { |
2167 | if (status == -ESTALE) { | |
2168 | nfs_zap_caches(inode); | |
2169 | if (!S_ISDIR(inode->i_mode)) | |
2170 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2171 | } | |
1da177e4 | 2172 | return status; |
a71ee337 | 2173 | } |
1da177e4 LT |
2174 | nfs_access_add_cache(inode, &cache); |
2175 | out: | |
e6305c43 | 2176 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2177 | return 0; |
2178 | return -EACCES; | |
2179 | } | |
2180 | ||
af22f94a TM |
2181 | static int nfs_open_permission_mask(int openflags) |
2182 | { | |
2183 | int mask = 0; | |
2184 | ||
2185 | if (openflags & FMODE_READ) | |
2186 | mask |= MAY_READ; | |
2187 | if (openflags & FMODE_WRITE) | |
2188 | mask |= MAY_WRITE; | |
2189 | if (openflags & FMODE_EXEC) | |
2190 | mask |= MAY_EXEC; | |
2191 | return mask; | |
2192 | } | |
2193 | ||
2194 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2195 | { | |
2196 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2197 | } | |
2198 | ||
e6305c43 | 2199 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2200 | { |
2201 | struct rpc_cred *cred; | |
2202 | int res = 0; | |
2203 | ||
91d5b470 CL |
2204 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2205 | ||
e6305c43 | 2206 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2207 | goto out; |
2208 | /* Is this sys_access() ? */ | |
9cfcac81 | 2209 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2210 | goto force_lookup; |
2211 | ||
2212 | switch (inode->i_mode & S_IFMT) { | |
2213 | case S_IFLNK: | |
2214 | goto out; | |
2215 | case S_IFREG: | |
2216 | /* NFSv4 has atomic_open... */ | |
2217 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2218 | && (mask & MAY_OPEN) |
2219 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2220 | goto out; |
2221 | break; | |
2222 | case S_IFDIR: | |
2223 | /* | |
2224 | * Optimize away all write operations, since the server | |
2225 | * will check permissions when we perform the op. | |
2226 | */ | |
2227 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2228 | goto out; | |
2229 | } | |
2230 | ||
2231 | force_lookup: | |
1da177e4 LT |
2232 | if (!NFS_PROTO(inode)->access) |
2233 | goto out_notsup; | |
2234 | ||
98a8e323 | 2235 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2236 | if (!IS_ERR(cred)) { |
2237 | res = nfs_do_access(inode, cred, mask); | |
2238 | put_rpccred(cred); | |
2239 | } else | |
2240 | res = PTR_ERR(cred); | |
1da177e4 | 2241 | out: |
f696a365 MS |
2242 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2243 | res = -EACCES; | |
2244 | ||
1e7cb3dc CL |
2245 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2246 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2247 | return res; |
2248 | out_notsup: | |
2249 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2250 | if (res == 0) | |
2251 | res = generic_permission(inode, mask, NULL); | |
1e7cb3dc | 2252 | goto out; |
1da177e4 LT |
2253 | } |
2254 | ||
2255 | /* | |
2256 | * Local variables: | |
2257 | * version-control: t | |
2258 | * kept-new-versions: 5 | |
2259 | * End: | |
2260 | */ |